A third kind of visitor
The shift is no longer hypothetical. OpenAI’s Dots and Meta’s Muse helped bring personal agents to a wider audience: Muse was downloaded 5 million times in its first month, while traffic from AI agents tripled from August to September.
Websites have not adapted at the same pace. Many block agents that users send to bargain or plan trips, in part because existing systems struggle to distinguish them from people and conventional bots. TechCrunch recently reported problems running agents on sites from Pizza Hut to Yelp; Yelp does not allow AI to book restaurant tables.
Amazon has blocked Muse agents and is suing Perplexity, seeking to bar access by its agents too. A judge recently ruled in Perplexity’s favor, but Amazon appealed. The company argues that agents threaten its advertising business by making it harder to tell whether ads are being shown to people or bots.
The obstacles are already frustrating early users. Ethan Mollick, an associate professor at the University of Pennsylvania’s Wharton School, expects the problem to spread across the internet. If agents start requesting refunds or negotiating prices at scale, customer-support teams could be overwhelmed, making them harder for people to use. Mollick says users may eventually resent being unable to get agents to handle these issues on their behalf.
There is a commercial reason for platforms to make room. E-commerce accounts for most agent traffic, and retailers could profit if agents are allowed to buy for their users. Shopify already accepts agents and gives them separate checkout windows that bypass pages designed for people.
The web agents leave behind
The risk is not simply that some websites will refuse agents. If platforms increasingly optimize for them, people who prefer to browse and compare products themselves could get a worse experience. Site verification might also become more intrusive: users could be asked repeatedly to prove they are human, perhaps even by scanning an iris.
Prices and recommendations could become harder to interpret, too. The same sweater might not have one price for everyone; the cost could vary according to whose agent asks. And users may have no way to tell whether an agent’s recommendation was shaped by advertising or a platform commission.
That would sit uneasily beside the promise made by Google, Anthropic and OpenAI: that agents will give consumers more choice and bargaining power. The promise is plausible. Agents could help people challenge unjustified medical bills or find cheaper car insurance. But whether they can do that depends on the rules governing how they interact with websites and what they are allowed to do.
Some rules should benefit everyone. Neither consumers nor companies want agents scattering credit-card details across the internet, so security requirements have a clear purpose. The harder disputes concern whether companies can shut agents out or stop them from negotiating better terms for buyers.
I think the central tension is not whether agents can make online tasks easier. It is who gets to set the terms when they do. A web that welcomes agents on platforms’ terms may make transactions simpler while leaving people with less visibility into the prices, pages and recommendations they encounter.
Daily AI news
Every day we pick what actually matters in AI and explain it plainly — no hype, no filler. Subscribe if you want to follow where the industry is going.
Only what matters — every day
Follow on X